First Canvas

First Canvas began the way real work tends to, quietly, out of necessity, after a filmmaker finally free to ask a question he'd carried for years. Ffish directed every frame, prompt, and performance cue himself, bringing decades of traditional production experience into tools most people were still learning to spell. The result feels authored rather than assembled. My contribution to the score sat closer to passing ingredients through the kitchen window than cooking the sidedish, with Ffish making the sound his own until the raw sonic elements carried the film's primal echo.

The score didn't arrive fully formed. Early prompts fed into ElevenLabs' Eleven Music engine kept misfiring into spoken word and chant whenever words like natural or echoing went in unanchored, the model reading voice-like language as an instruction to sing rather than to play. Getting the ceremonial horn right meant stripping anything that could read as vocal, tribal, ancient, ceremonial, until the prompt pinned itself to a physical sound source and a performance style instead of a mood. Stone percussion needed the same correction, an explicit refusal of drums, language built around rattling, clattering, dry and irregular rhythm, so the engine understood stones as stones rather than a drum kit standing in for them.

ChatGPT became a research partner mid-process, mapping tone references, Moondog's street-primal minimalism, Don Cherry's bone-horns and communal scat, Paul Horn's sacred space resonance, against what the film's prehistoric world actually needed sonically. That research shaped a loose internal direction nicknamed Caveman Jazz, ritual and improvisation standing in for melody.

Raw generations moved through GarageBand and Adobe Audition for sweetening before landing in Ffish's timeline, where his editorial hand folded them into the finished sound design. What survives in the film carries all of that trial and error underneath it, even though none of it shows on the surface.